Discovery Ads: real incremental 2026?

58 to 72% of Discovery accounts overestimate their ROAS by +18 to +46% solely due to Google's view-through attribution window. The median over-attribution ratio observed on 2,000+ accounts: 2.5 to 3.5x — a displayed 4x ROAS often corresponds to a real incremental ROAS of 1.2 to 1.8x.

Discovery Ads ticks all the boxes of an attractive format: premium placements (YouTube Home, Gmail Promotions, Discover feed), CPM -20 to -32% cheaper than equivalent YouTube ads depending on vertical, displayed ROAS often spectacular (3 to 5x in the Google Ads UI). The measured reality is less glorious. On our 2025 sector benchmark, 58 to 72% of Discovery accounts overestimate their ROAS by +18 to +46%, solely due to a generous view-through attribution window.

This guide isn't an anti-Discovery plea. It's a critical audit: where Discovery really pays, where it's a money pit, how to measure real incrementality (three proven methods), how to configure it cleanly in 2026 after the Demand Gen migration. We also cover the boundary with Performance Max and budget trade-offs between formats. Goal: that by the end, you know whether Discovery deserves your $3,000/month, or whether you need to reallocate it elsewhere.

What are Discovery Ads in 2026 and what changed with Demand Gen?

Discovery Ads launched in 2019 by Google as a native visual format positioned on three premium inventories: the Discover feed (Google app, Android home), YouTube Home (vertical scroll, pre-click), and the Gmail Promotions tab. The initial promise: offer image/carousel format on placements where users are in discovery mode, without explicit Search intent, but with a very rich YouTube-Google behavioral signal. Key difference vs the classic Display network: unified premium placements rather than a long tail of third-party publishers, and audiences boosted by YouTube signals.

What changed in late 2023 — 2024: Google rebranded Discovery as Demand Gen Ads and integrated new placements (YouTube Shorts, short YouTube in-stream, YouTube in-feed). The bidding algorithm and attribution logic remain similar, but Google improved lookalike segments (similar audiences rebuilt on first-party signal), added a clearer per-placement reporting breakdown, and unified mobile inventory. All Discovery accounts were automatically migrated in early 2024. Official documentation on Google Ads Support.

In practice, Demand Gen in 2026 targets three uses:

• Top-mid funnel acquisition on warm audiences (combined lookalike, in-market, affinity).
• Enhanced visual remarketing on recent site visitors + customer match — often better ROI than classic Display remarketing.
• Product launch / seasonal campaign: strong visual formats, push a message for 4 to 8 weeks, cut once the season passes.

What you shouldn't expect from Discovery: purely demand-capture conversions. Demand Gen generates demand (as its name now indicates), it doesn't capture bottom-funnel intent. Confusing these two roles is the main cause of poor ROAS evaluation — detailed in section 2.

Why does Google over-attribute Discovery conversions so aggressively?

Open the interface. Your Discovery campaign shows 280 conversions over 30 days, ROAS 4.1x. Immediate temptation: double the budget. Mistake. A significant portion of these conversions would have happened without Discovery. They're simply retouched by a view-through impression and Google credits them with particular generosity on this format. This is the view-through over-attribution phenomenon, particularly marked on Discovery because of its default 3-30d attribution window.

On accounts we continuously monitor, median Discovery incrementality measured via geographic holdout sits between +5 and +14% truly incremental conversions, where Google attributes +30% conversions to the campaign. In other words, when the UI says "Discovery brought 300 conv," clean measurement says 70 to 120 incremental conv. Median observed over-attribution ratio: 2.5 to 3.5x.

Three concrete causes of Discovery over-attribution:

• View-through conversions counted up to 30 days after viewed (not clicked) impression — particularly high on YouTube Home and Discover feed.
• Engaged-view conversions (YouTube): 10 seconds of viewing is enough to claim the future conversion, even without click or interaction.
• Aggressive cross-device attribution: a user sees the ad on mobile Gmail, converts 9 days later on desktop via Search — Discovery gets credited.

If you've never done a geo holdout and Discovery represents more than 20% of your budget, you are statistically very likely over-allocating. To dig into attribution logic, see our ROAS / CPA / CPC guide.

What are the 3 Discovery placements (YouTube Home, Gmail, Discover feed)?

Demand Gen doesn't display uniformly: each placement has a distinct audience, CVR, and conversion profile. Knowing them lets you orient your creatives and exclusions. In our sector panel, observed conversion rates are noticeably different across placements.

• YouTube Home (CVR 2.3%) — vertical scroll on the YouTube homepage, the most engaging placement. Audience in video discovery mode, rich YouTube behavioral signal. More qualified conversions when creatives are visual and short. It's the placement to prioritize if your videos are quality.
• Gmail Promotions (CVR 1.8%) — Promotions tab. Audience generally end-of-funnel CRM (repeat opens, historical Google engagement). High open rate but fuzzy view-through attribution: difficult to distinguish real incrementality from passive reading. Exclude for restricted budgets.
• Discover feed (CVR 1.5%) — Discover feed in the Google app and Android home. Broadest audience, least qualified. Huge volume but lowest CVR. Useful for top-funnel reach campaigns, not for direct conversion. Exclude it if your Discovery CPA exceeds your target by 20% or more.

Concrete action: in the 2026 Demand Gen reporting, Google now exposes a per-placement breakdown (under "Serving details"). Check after 30 days what share of your spend goes on each placement. If Discover feed absorbs 60% of the budget with a 1.5% CVR, there's probably rebalancing to do — via placement exclusions or creative adjustments.

When does Discovery Ads really work?

Discovery (Demand Gen) isn't a universal format. Depending on vertical, around 28 to 42% of accounts achieve a measured incremental ROAS > 1.5x — the threshold we consider net profitable. The five winning contexts share: strong visuals, short conversion cycle, and an audience to activate.

• E-commerce fashion / beauty / lifestyle — strong product visuals, native visual shopping, short decision cycle. Typical measured incremental ROAS: 2.5 to 4x. Discovery amplifies an already desirable catalog.
• DTC brands with storytelling — good 15-second video narration + image carousel. Discovery rewards creative quality more than any other Google format. A brand with a solid visual identity outperforms.
• Mobile apps with strong visual hook (games, fintech, fitness, food delivery). The YouTube Shorts Demand Gen placement opened in 2024 is particularly effective for mobile install.
• Time-limited events — sales, product launches, promo weeks. Discovery excels on 2-to-8-week windows where urgency pushes conversion and the algorithm doesn't need 90 days to learn.
• Upsell / cross-sell on existing customer base — Customer Match + 90d visitor remarketing. Incremental ROAS often > 3x because the audience is already familiar with the brand; true incrementality vs email/SMS base is the real benchmark to test.

Common trait: Discovery works when the brand has something to show. If your creatives are weak, no audience will change that. Invest 20 to 30% of Discovery budget in creative production before investing in media — it's the only variable that moves CVR on this format.

When is Discovery a budget money pit?

Symmetrically, in our sector panel, around 34 to 46% of Discovery accounts show a median incremental ROAS between 0.4 and 0.9x — purely destructive. In these cases, every dollar invested in Discovery destroys value: real CPA (excluding inflated attribution) is often +32 to +48% vs equivalent Search. The five losing contexts are easy to recognize.

• B2B services with no precise intent — consulting, advisory firms, financial services. Discovery touches a discovery-mode audience that's never actively searching for a B2B provider. Typical incremental ROAS: 0.3 to 0.6x. Prefer Search + LinkedIn.
• Pure SaaS with a conversion cycle > 30 days. The Discovery attribution window (max 30d) doesn't cover long B2B decisions. View-through seen but not converted exit the cookie before the sale — no credit, very low displayed ROAS, and little real incrementality.
• Commoditized products with no visual differentiation — commodities, spare parts, consumables. Discovery is a visual format. If your product looks like that of 12 competitors, creatives create no advantage.
• Budgets < $2,000/month — Demand Gen needs 50 conversions over 14 days to exit learning phase. Below $2,000, you stay stuck in exploration for 3 to 6 weeks without ever stabilizing. Eternal learning phase = guaranteed money pit.
• Accounts with no prior remarketing (fully cold audience) — Discovery works 3x better with first-party signals (site visitors, customer match). Starting from scratch on pure cold lookalike audience: median incremental ROAS 0.4x, budget burned in exploration.

If your account meets 2 of these 5 criteria, Discovery is probably to cut or not to launch. Reallocated to Search + Shopping, the same budget produces on median 1.6 to 2.1x more incremental conversions depending on vertical, on accounts that made the reallocation.

How do you measure Discovery incrementality with 3 methods?

The ROAS displayed in the Google Ads UI isn't the incremental. To make a clean decision, three proven methods — with different cost / time / reliability profiles. Choose based on your budget and data maturity.

1. Geographic holdout — the universal method, accessible to all. Cut Discovery on 2 representative regions (15 to 25% of volume) for 28 consecutive days. Measure total conversions across all campaigns on test zones vs control zones. The difference is your incremental. Procedure detailed in the HowTo JSON-LD of this article. Real cost: 15 to 25% opportunity cost for 4 weeks — generally paid back by the informed decision that results.

2. Official Google lift test — Conversion Lift service provided by Google for accounts > $25,000/month. A Google Account Manager sets up a clean geographic or user-based A/B test with robust statistical control. Documentation on Google Ads Support. Advantage: Google guarantees statistical power. Drawback: high spend threshold, opaque configuration. Relevant for validating an important strategic trade-off.

3. Bayesian Marketing Mix Model (MMM) — for large multi-channel budgets. An MMM estimates the incremental contribution of each channel (Search, Discovery, Meta, YouTube, TV, offline) on total sales, accounting for seasonality, halo effects, and latencies. Useful read: Think with Google's analysis on the evolution of modern MMM models. Setup cost $8 to $20k, then $2 to $5k/month maintenance. Relevant threshold: marketing budget > $100k/month.

For the majority of accounts (budget $5 to $50k/month Discovery), geo holdout is the default choice. Simple, proven, results in 4 weeks. Our SteerAds audit module includes automatic detection of the likely incrementality ratio before launching a real holdout — useful to prioritize which accounts to test first.

How do you configure an optimal Discovery setup?

If you're in a winning case (section 4), here's the Discovery / Demand Gen setup that maximizes the chances of an incremental ROAS > 1.5x. Three blocks: audiences, creatives, budget and bidding.

Audiences — combine 4 sources:

• Customer Match — existing customers uploaded by hashed email (5,000+ contacts minimum). Most powerful signal, essential for upsell.
• Lookalike / similar audiences — automatically built from your customer match. Main acquisition gate.
• Google In-market — thematic audiences (e.g., "Women's clothing," "Management software"). 2 to 4 segments per campaign.
• Affinity — broader, to prime top-funnel when lookalike volume is limited.

Creatives — the minimums to exit learning:

• 15 images (1:1 square, 1.91:1 landscape, 4:5 portrait) — varied formats for all placements.
• 5 short videos 6 to 15 seconds, ideally 9:16 vertical for YouTube Shorts.
• 5 short headlines < 40 characters — the headline carries the offer.
• 5 descriptions with strong CTA ("Try," "Buy," "Get").
• 2 to 3 HD logos on transparent background.

Budget and bidding:

• Absolute floor $2,000/month, recommended $3,000 to $5,000/month for a real test.
• Bid strategy D1-D14: Maximize Conversions (not Target CPA). Let the algorithm explore freely in the first learning phase.
• D15 — switch to Target CPA at -10% of the average CPA observed in the first 14 days. No more aggressive: you'd constrain the algorithm.
• D30 — adjustment tier: budget scaling +20% max if ROAS stable, then retest. Doubling the budget at once resets the learning phase.

For all the conversion tracking logic supporting this measurement, see our Google Ads conversion tracking guide. And for the broader e-commerce playbook orchestrating Discovery with Search, Shopping, and PMax, our 2026 e-commerce playbook.

How do you manage the Discovery to Demand Gen Ads migration?

In late 2023, Google announced the Discovery Ads → Demand Gen Ads rebranding. The migration of all accounts happened automatically between February and June 2024. For most advertisers, the change was transparent — same interface, same bidding logic. But three practical differences have real implications that few articles cover well.

1. YouTube Shorts + short in-stream integration. Discovery didn't serve on YouTube Shorts. Demand Gen does, and these placements now represent 18 to 28% of impressions of a typical Demand Gen campaign (median, variable by vertical). Impact: 9:16 vertical formats become critical. If you don't have vertical creative, you're under-leveraging the best-performing placement in 2026.

2. Improved lookalike segments. Demand Gen now offers similar segments rebuilt on enriched first-party signal (customer match + enhanced conversions + Google signals). "Narrow" quality lookalikes (top 1% most similar) are significantly more performant than in classic Discovery — CVR +18 to +28% depending on vertical. Check that your similar audiences have been regenerated in Demand Gen mode.

3. Per-placement reporting breakdown. Demand Gen finally exposes a serving report by placement (YouTube Home, Gmail, Discover feed, YouTube Shorts), something Discovery largely hid. This enables targeted exclusions from section 3 and informed decision-making on where to concentrate budget. If you haven't yet consulted this breakdown in your account, it's the first action to take after reading this guide.

Overall measured impact: on accounts we continuously monitor post-migration, attributed ROAS +3 to +7% at median. Important: this rise is largely driven by the addition of more relevant placements (Shorts), not by a real improvement in underlying incrementality. When we re-measure via post-migration holdout, the real incrementality gain is more modest: +1 to +3%.

To compare this Demand Gen audit with the Performance Max audit (the two unified formats share similar attribution patterns), see our complete 2026 Performance Max guide. Incrementality measurement methods apply identically. And to industrialize continuous monitoring of these campaigns, our auto-optimization module monitors attribution drifts in real time. Complementary official resource to contextualize Demand Gen within Google's offering: the official product page.

Sources

Official sources consulted for this guide:

• support.google.com
• support.google.com
• support.google.com

FAQ